JPH0937255A - Motion parameter detector, motion parameter detection method and image coder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To calculate an accurate motion parameter with less calculation amount. SOLUTION: A reliability point calculation circuit 41 selects two prescribed points from points of a chain in an image and detects respective motion vectors of the two points. A motion parameter calculation circuit 42 obtains motion parameters representing respectively a chain translation amount (Vx, Vy), a magnification reduction factor S and a rotation amount R, as equations: (Vx, Vy)=((Vx1+Vx2)/2, (Vy1+Vy2)/2), S=((1-F)<2> +G<2> )<1/2> , and R=arctan(G/(1-F)), where (x1, y1),(x2, y2) are coordinates of two points, (Vx1, Vy1), (Vx2, Vy2) are respective motion vectors, F=(AC+BD)/E, G=(AD-BC)/E, A=Vx1-Vx2, B=Vy1-Vy2, C=x2-x1, D=y2-y1, and E=C<2> +D<2> .

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion parameter detecting device, a motion parameter detecting method, and an image coding device. In particular, the present invention relates to a motion parameter detection device and a motion parameter detection method, and an image coding device capable of accurately detecting a motion parameter representing the motion of continuous points such as a chain connecting feature points of an image. .

[0002]

2. Description of the Related Art Conventionally, in an apparatus for transmitting an image under a limited transmission band or an apparatus for recording on a recording medium having a limited storage capacity, the amount of image data is reduced. A high-efficiency coding method for efficiently compressing with code words is adopted. As a high-efficiency image coding method, for example, an input image is orthogonally transformed by DCT (discrete cosine transform) and adaptive quantization is performed for each frequency band according to human visual characteristics, and a wavelet basis ( A method is known in which an image is divided into subbands by wavelet transformation, and each band is weighted and encoded. According to these encoding methods, distortion is visually inconspicuous, and
A high compression rate can be realized.

However, in these encoding systems, if the compression ratio is further improved, the visually unfavorable effects such as so-called block distortion become remarkable.

Therefore, as an encoding method in which visually harmful distortion does not occur even at a high compression rate, for example, characteristic points (characteristic points) of the structure of an image (for example, points (pixels) forming the outline of an object, etc.) ) Is extracted and the image data at that feature point is efficiently encoded, and a structure extraction encoding method by image feature point detection is known.

FIG. 37 shows the configuration of an example of a conventional image coding apparatus for compressing and coding an image by the structure extraction coding method. The image data is input to the quantizer 11 and the two-dimensional change point detection circuit 10. In the quantizer 11, the image data is quantized, and the quantized coefficient is obtained. This quantized coefficient is supplied to the chain (chain) encoding circuit 12. Further, the two-dimensional change point detection circuit 10 detects whether or not the input image data is a feature point, and if the input image data is a feature point, for example, the feature point data having a value of 1 is If it is not a feature point, feature point data having a value of 0 is output to the chain encoding circuit 12, respectively.

When the quantization process in the quantizer 11 and the feature point detection process in the two-dimensional change point detection circuit 10 are completed for one frame of image data,
In the chain encoding circuit 12, the position information (position (coordinates) of the feature point) about the point (pixel) having the feature point data of 1 is chain-encoded, and further the quantized coefficient of the feature point (detected as the feature point is detected. It is multiplexed with the quantized coefficient of the image data in the pixel) to form chain encoded data. The chain-encoded data is supplied to the buffer (buffer memory) 17, is temporarily stored, and then recorded on a recording medium (not shown) or output to the transmission path. Note that the chain-encoded data is temporarily stored in the buffer 17 so that the amount of data is smoothed,
As a result, the chain encoded data is output to the recording medium or the transmission path at a substantially constant data rate.

[0007]

By the way, since the above-mentioned image coding apparatus targets a still image for coding, when a moving image is coded by this image coding apparatus, the moving picture in the time direction is The compression rate is poor as compared with a transform coding method for removing redundant components, which includes, for example, motion compensation processing and DCT processing. Therefore, when chain coding is performed on a moving image, a method for improving the compression rate is, for example, by associating chains having the same contour with each other over a plurality of frames and by indicating a motion indicating the motion of the chain. The applicant of the present application has previously proposed a method in which a parameter (motion vector) is detected and any one of the chain-encoded data in the associated chains and the detected motion parameter are multiplexed and output.

FIG. 38 shows an example of the configuration of a motion parameter detecting device (motion vector detecting device) adopted in such an image coding device or a motion tracking device for tracking the movement of the contour of an object. Shows. The chain coordinate generation circuit 311 detects a chain corresponding to the chain coded data based on a direction component (described later) of the chain included in the chain coded data output from the chain coding circuit 12, and further detects the chain. Find the coordinates of each feature point that makes up the chain. Then, the chain coordinate generation circuit 311 sequentially outputs the obtained coordinates as the current frame coordinates to the difference calculation circuit 314.

Further, the chain coordinate generation circuit 311 uses the affine parameters while changing each coefficient of, for example, affine parameters, which is a motion parameter representing the motion of the image, within a certain range, and uses the affine parameters as the coordinates of the current frame. Of the difference calculation circuit 314.
And the affine parameters used at that time are output to the selector 315 as motion parameters.

That is, the coefficient of the affine parameter is a,
b, c, d, e, f, and the current frame coordinates or the subsequent frame coordinates are (x, y) or (x ', y'), respectively, the chain coordinate generation circuit 31
1 is changed according to the following equation, for example, by changing the coefficients a, b, d, and e in the range of −2 to 2, and the coefficients c or f in the range of −8 to 8, respectively, in appropriate steps, and The frame coordinates (x ', y') are calculated. x '= ax + by + cy' = dx + ey + f

Here, the above equation is generally expressed as shown in FIG.

Therefore, from the chain coordinate generation circuit 311 to the difference calculation circuit 314, for one current frame coordinate, the current frame coordinate is converted by a plurality of affine parameters, that is, a post frame coordinate, that is, a plurality of post frame. Coordinates are output, and the chain coordinate generation circuit 311 outputs a plurality of affine parameters to the selector 315.

On the other hand, the current frame buffer 312 is supplied with image data of a frame in which a chain whose motion parameter is to be detected is present. ,
The current frame data) is stored. Further, the rear frame buffer 313 is supplied with image data of the next frame (rear frame) of the current frame data, in which the image data (hereinafter, appropriately,
The subsequent frame data) is stored.

The difference calculation circuit 314 receives the current frame coordinates and the subsequent frame coordinates from the chain coordinate generation circuit 311 after the current frame data or the subsequent frame data is stored in the current frame buffer 312 or the subsequent frame buffer 313, respectively. Block matching is performed on blocks within a predetermined range centered on the current frame coordinates or the subsequent frame coordinates.

That is, the difference calculation circuit 314 outputs the current frame coordinate or the subsequent frame coordinate received from the chain coordinate generation circuit 311 as an address to the current frame buffer 312 or the subsequent frame buffer 313, respectively. 312 or the rear frame buffer 313, Q × Q (horizontal ×) centered on the pixel corresponding to the current frame coordinate or the rear frame coordinate.
The current frame data or the subsequent frame data (pixel value) in the range of (vertical) pixels is read out.

Then, the difference calculation circuit 314 determines each of the current frame data in the range of Q × Q pixels centered on the pixel corresponding to the current frame coordinates and the Q × Q pixels centered on the pixel corresponding to the subsequent frame coordinates. The difference with each of the subsequent frame data in the range (1) is obtained (hereinafter, this difference value is appropriately referred to as a prediction error), and further, for example, the sum of absolute values of the prediction error is calculated.

The difference calculation circuit 314 obtains the sum of absolute values of the prediction errors as described above for each of a plurality of subsequent frame coordinates obtained for one current frame coordinate and outputs it to the selector 315.

Before the chain coordinate generation circuit 311 starts processing for a chain, the selector 315 starts
The minimum error register 316 is initialized by setting -1 in the minimum error register 316, for example. And
When the selector 315 receives the sum of absolute values of prediction errors from the difference calculation circuit 314, the selector 315 and the minimum error register 3
16 if the sum of absolute values of prediction errors is smaller than the storage value of the minimum error register 316, or if the storage value of the first error register 316 is −1, the prediction error from the difference calculation circuit 314 is calculated. The sum of absolute values is newly set in the minimum error register 316. Furthermore, the selector 315
Causes the motion parameter register 317 to store the affine parameters supplied from the chain coordinate generation circuit 311 and used to obtain the sum of the absolute values of the prediction errors set in the minimum error register 316.

Then, the chain coordinate generation circuit 311
When the process for one chain is completed, the motion parameter output circuit 318 reads the affine parameter stored in the motion parameter register 317 at that time and outputs it as the motion parameter.

As described above, in the motion parameter detecting device shown in FIG. 38, the coordinates are converted by the affine parameters of various values, and the coordinates before conversion are Q × Q.
The sum of absolute values of the difference between the current frame data of the range and the post frame data of the range of Q × Q centered on the coordinates after conversion, that is, the sum of the absolute values of the prediction errors, The affine parameter that minimizes the sum of the absolute values of the prediction errors is detected as the motion parameter.

Therefore, there is a problem that an erroneous motion parameter is detected.

That is, for example, when a linear chain CH _A is present in the current frame as shown in FIG. 40 (A), as shown in FIG. Frame), and the chain is in a non-moving state,
(C), the chain CH _A is, or motion parameters are detected that represent the translation state along the chain CH _A, or, as shown in FIG. 40 (D), the chain CH _A is reduced In some cases, a motion parameter that represents the generated state has been detected.

Further, for example, when a V-shaped chain CH _B exists in the current frame as shown in FIG. 41 (A), the chain is changed in the rear frame as shown in FIG. 41 (B). However, when the position of the apex is in the expanded state without moving,
As shown in (C), a motion parameter representing a state in which the chain CH _B is translated may be detected.

Further, according to the motion vector detecting device of FIG. 38, since the motion parameter is detected by using the affine parameter having the above-mentioned six coefficients a to f, the parallel movement of the object, the scaling, And motion parameters representing motion such as rotation can be obtained. However, in this case, since as many as six coefficients of the coefficients a to f are changed within the above-described range to calculate the prediction error, a huge amount of calculation is required, and therefore the real-time processing is performed. In order to perform the above, it is necessary to use hardware capable of high-speed processing, and there has been a problem that the cost of the device increases.

The present invention has been made in view of such a situation, and makes it possible to detect an accurate motion parameter quickly.

[0026]

A motion parameter detecting apparatus according to the present invention provides a selecting means for selecting one or more points among continuous points and a motion vector for each of the one or more points selected by the selecting means. It is characterized by comprising motion vector detecting means for detecting and motion parameter calculating means for calculating motion parameters of consecutive points based on the motion vector detected by the motion vector detecting means.

In this motion parameter detecting device,
The selection means is caused to select two points out of the continuous points, and the motion parameter calculation means is based on the motion vector for each of the two points, the parallel movement amount of the continuous points, the enlargement / reduction ratio,
Alternatively, a motion parameter representing the amount of rotation can be calculated. Further, the selecting means can calculate the reliability representing the reliability of each continuous point, and select two points based on the reliability. Furthermore, the reliability of each continuous point can be calculated by the selecting means by performing block matching using a block in a predetermined range centered on that point, and calculating the reliability based on the prediction error obtained at that time. . Furthermore, the selecting means can calculate the reliability of each continuous point based on the angle formed by the continuous points with the point as the vertex. Further, it is possible to cause the selection means to select a continuous point that satisfies a predetermined condition and has the highest reliability and the next highest point. Further, the selecting means can select the start point and the end point of the continuous points when no two or more continuous points satisfy the predetermined condition. Further, the selection means can select the start point and the end point of the continuous points.

In the motion parameter calculating means, the average value of the motion vectors detected by the motion vector detecting means is
It can be calculated as a motion parameter indicating the parallel movement amount of the continuous points. Further, in this case, the selecting means can be made to calculate the reliability representing the reliability of each continuous point and select one or more of the continuous points based on the reliability.

According to the motion parameter detecting method of the present invention, one or more points are selected from the continuous points, the motion vector for each of the one or more points is detected, and the motion of the continuous points is calculated based on the motion vector. It is characterized by calculating parameters.

In the image coding apparatus of the present invention, the motion parameter detecting means for detecting a motion parameter which is a parameter representing the motion of the chain connecting the feature points selects one or more of the feature points forming the chain. Selection means to
A motion vector detecting means for detecting a motion vector for each of the one or more points selected by the selecting means, and a motion parameter calculating means for calculating a motion parameter of the chain based on the motion vector detected by the motion vector detecting means. It is characterized by including.

In the motion parameter detecting device of the present invention, the selecting means selects one or more points among the continuous points,
The motion vector detecting means is 1 selected by the selecting means.
The motion vector for each of the above points is detected. The motion parameter calculation means is configured to calculate the motion parameter of the continuous points based on the motion vector detected by the motion vector detection means.

In the motion parameter detecting method of the present invention, one or more points among the continuous points are selected, the motion vector for each of the one or more points is detected, and the continuous point is detected based on the motion vector. The motion parameters are calculated.

In the image coding apparatus of the present invention, the selecting means selects one or more points out of the characteristic points forming the chain, and the motion vector detecting means the one or more points selected by the selecting means. The motion vector for each is detected. The motion parameter calculation means is configured to calculate the motion parameter of the chain based on the motion vector detected by the motion vector detection means.

[0034]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below, but before that, in order to clarify the correspondence between each means of the invention described in the claims and the following embodiments. The features of the present invention are described as follows by adding a corresponding embodiment (however, an example) in parentheses after each means.

That is, the motion parameter detection device according to the first aspect is a motion parameter detection device that detects a motion parameter that is a parameter representing the motion of a continuous point in an image, and one or more of the continuous points. Selection means for selecting points (for example, the confidence point calculation circuit 41 shown in FIG. 29),
Based on the motion vector detecting means (for example, the confidence point calculating circuit 41 shown in FIG. 29) for detecting the motion vector for each of the one or more points selected by the selecting means, and the motion vector detected by the motion vector detecting means. And a motion parameter calculation means (for example, the motion parameter calculation circuit 42 shown in FIG. 29) for calculating the motion parameters of the continuous points.

An image coding apparatus according to a twelfth aspect of the present invention is a feature point coding means (for example, the chain coding circuit 12 shown in FIG. 1 or the like) that codes information relating to feature points of a moving image and outputs coded data. ), A motion parameter detecting means (for example, a motion parameter calculating circuit 125 shown in FIG. 17) that detects a motion parameter that is a parameter representing the motion of the chain connecting the feature points, and the code output from the feature point coding means. Multiplexing means for multiplexing the encoded data and the motion parameter detected by the motion parameter detecting means (for example, the select multiplexing circuit 16 shown in FIG. 1).
An image encoding apparatus having: and a motion parameter detecting means, and a selecting means (for example, a confidence point calculating circuit 41 shown in FIG. 29) for selecting one or more points among the characteristic points forming the chain. , The motion vector detecting means (for example, the confidence point calculating circuit 41 shown in FIG. 29) for detecting the motion vector for each of the one or more points selected by the selecting means, and the motion vector detected by the motion vector detecting means. On the basis of this, there is provided a motion parameter calculating means for calculating the motion parameter of the chain (for example, the motion parameter calculating circuit 42 shown in FIG. 29).

Of course, this description does not mean that each means is limited to those described above.

FIG. 1 is a block diagram showing the configuration of a first embodiment of an image coding apparatus to which the present invention is applied. Note that, in the figure, portions corresponding to those in FIG. 37 are denoted by the same reference numerals.

This image coding apparatus is adapted to compress and code a moving image by the structure extraction coding method. That is, for example, the input image data obtained by digitizing the image signal of the moving image is passed through the two-dimensional change point detection circuit 10 or the quantizer 11 as described with reference to FIG.
By being supplied to the chain encoding circuit 12, it becomes chain encoded data. This chain encoded data is
It is supplied to the selector 13. The selector 13 stores the chain-encoded data in units of frames in a RAM (Chain Code).
RAM) 14 _{1 to} 14 _N.

That is, the chain replacing circuit 15 and the select multiplexing circuit 16 in the subsequent stage are in units of N frames (for example,
The processing is performed in units of 15 frames), and the _first frame (first frame) of the chain encoded data in units of N frames is stored in the RAM 14 ₁ . Stores the chain encoded data of the second frame, and thereafter, similarly stores the chain encoded data of the nth frame in the RAM 14 _n (however, n = 1, 2, ...,
N).

Here, the selector 13, when storing the chain encoded data in the RAM 14, adds a chain number as a unique number to each chain encoded data (each chain). There is. In the image encoding device of FIG. 1, when handling a chain (chain encoded data), this chain number is
Used to identify each chain. The chain number may be unique for the chains existing in each frame, but the chain replacement circuit 15
All chains existing in the N frame, which is the processing unit of, may be unique.

On the other hand, in this image coding apparatus, the image data is also supplied to the selector 19. The selector 19 stores the image data in the RAMs 20 _{1 to} 20 _N in frame units. That is, thereby, the RAM 20 _n, as in RAM 14 _n, the image data of the n-th frame is stored.

In the RAMs 14 _{1 to} 14 _N , chain-coded data for N frames are stored, and the RAM 2
0 to ₁ to 20 _N, the image data of N frames are stored, the chain substitution circuit 15, is read out as necessary chain coded data stored in the RAM 14 _n. Then, present in the first frame, there chain (thus, stored in RAM 14 _1, there chain)
, The motion parameter to the subsequent frame (motion parameter for the subsequent frame) is detected. The chain replacement circuit 15 is adapted to detect a motion parameter based on the image data stored in the RAM 20.

Furthermore, the chain moves according to the detected motion parameters (not only translation but also rotation, scaling, etc. (thus, to be exact,
(Translation)), and the importance (Visual Significa) that represents the visual importance of the post-movement chain (hereinafter referred to as the "movement chain") in the subsequent frame.
nt) is calculated.

Then, when the importance is equal to or higher than the predetermined threshold value T, the chain replacement circuit 15 replaces the chain around the moving chain among the chains existing in the rear frame with the moving chain, and after the replacement, The same process is performed on the subsequent frames using the chain until the importance does not exceed the threshold value T or until the Nth frame. After that, the chain replacement circuit 15 determines the chain number of the chain of the first frame used first, the motion parameter to the subsequent frame from the first frame to the last frame where the chain has been replaced, and the number of the motion parameters. The (number of frames from the first frame to the last frame where chain replacement has been performed) is output to the select multiplexing circuit 16.

The select multiplexing circuit 16, the chain coded data for the chain corresponding to the chain number of the chain substitution circuit 15 reads from the RAM 14 _1, the chain coded data, and motion parameters from the chain substitution circuit 15 And multiplexes the number of them and outputs as multiplexed chain encoded data.
This multiplexed chain encoded data is output to the transmission path via the buffer 17. Alternatively, for example, a recording medium 18 such as a magnetic tape, an optical disc, or a magneto-optical disc.
Will be supplied to and recorded.

FIG. 2 shows a configuration example of the chain encoding circuit 12 of FIG. The quantized coefficient from the quantizer 11 is supplied to and stored in the coefficient frame buffer 21. In the coefficient frame buffer 21, the quantization coefficient Q _ij of the image data at the point (i, j) on the image is represented by i or j, respectively, as shown in FIG. (Where i and j represent, for example, the horizontal (x-axis direction) or horizontal (y-axis direction) coordinates of pixels (points) forming an image) .

The feature point data from the two-dimensional change point detection circuit 10 is supplied to the mask frame buffer 22 and stored therein. Note that the mask frame buffer 22 also stores the feature point data in the same manner as in the coefficient frame buffer 21 described above.

When one frame of quantized coefficients or feature point data is stored in the coefficient frame buffer 21 or the mask frame buffer 22, respectively, the direction searcher 23 performs the processing according to the flowchart shown in FIG. Be seen.

That is, first, in step S1, the status register 24, the X coordinate register 25, and the Y register.
The coordinate register 26 is initialized to 0, and the process proceeds to step S2 to determine whether the stored value of the state register 24 is 0 or not. When it is determined in step S2 that the value stored in the state register 24 is 0, the process proceeds to step S3, and it is determined whether there are unprocessed coordinates. That is,
The direction searcher 23 determines X in step S4 described later.
In the coordinate register 25 or the Y coordinate register 26, x or y coordinates of pixels forming one frame image are set in, for example, a so-called line scan order (from left to right and from top to bottom). In step S3, it is determined whether the x or y coordinate of the pixel located at the end when the image is viewed in the line scan order is set in the X coordinate register 25 or the Y coordinate register 26 (for example, When the image of one frame consists of 640 × 400 pixels, the X coordinate register 25 or the Y coordinate register 26 has 640
(639) or 400 (399) is set).

When it is determined in step S3 that there is an unprocessed coordinate, the process proceeds to step S4, and the unprocessed coordinate (x, y) that appears first in the line scan order of the image is displayed.
(Coordinates) are set in the X coordinate register 25 and the Y coordinate register 26. Therefore, immediately after the quantized coefficient for one frame or the feature point data is stored in the coefficient frame buffer 21 or the mask frame buffer 22, 0 is set in both the X coordinate register 25 and the Y coordinate register 26. It

Here, in the chain encoding circuit 12, X
The pixel of the coordinates corresponding to the stored values of the coordinate register 25 and the Y coordinate register 26 is the target of the process first, but the pixel which is (is) the target of the process is hereinafter referred to as the pixel of interest as appropriate. Say.

Thereafter, the process proceeds to step S5, where the feature point data corresponding to the pixel of interest is mask frame buffer 22.
Is read out (the feature point data is read out from the address having the x-coordinate or the y-coordinate of the pixel of interest as the upper address or the lower address, respectively), and it is determined whether it is 1. If it is determined in step S5 that the feature point data is not 1 (if it is determined to be 0), that is, if the pixel of interest is not a feature point, the process returns to step S2. If it is determined in step S5 that the feature point data is 1, that is, if the pixel of interest is a feature point, the process proceeds to step S6, and the search X coordinate register 27 or the search Y coordinate register 28 stores the X coordinate. The value stored in the register 25 or the Y coordinate register 26 is set. Further, in step S6, the 3-bit counter 29 is initialized to 0. The 3-bit counter 29 can count a number that can be represented by 3 bits, that is, 0 to 7 (2 ³ −1), for example.

Then, in the direction searching device 23, step S
At 7, the valid data selection signal whose value is 1 (H level) is output. The valid data selection signal is usually
It is set to 0 (L level). Further, the valid data selection signal is supplied to the latch circuit 34, the selector 36, and the multiplexer 37. further,
The valid data selection signal is set to 0 again when the time for one clock elapses after it is changed from 0 to 1.

Further, in step S7, the coordinates stored in the search X coordinate register 27 and the search Y coordinate register 28 are read out, and the selector 3 is set as the start point coordinate.
6 is output. In step S7, 1 is set in the status register 24, and the feature point data having a value of 1 and corresponding to the pixel of interest stored in the mask frame buffer 22 is rewritten to 0.
Return to 2.

Here, the selector 36 has a direction searching device 2
3, the valid data selection signal and the start point coordinates are supplied, the stored value of the state register 24, the ROM (Direction
The output of the ROM 33 and the output of the direction change signal generator 35 are supplied. When the selector 36 receives the valid data selection signal whose value is 1, the selector 36 refers to the value stored in the state register 24, and when it is 0, the start point coordinate from the direction searcher 23 and the output of the ROM 33. , Or the start point coordinate from the direction searching device 23 among the outputs of the direction change signal generator 35, and outputs it to the multiplexer 37.

Further, when the multiplexer 37 receives the valid data selection signal having the value 1, the multiplexer 37 reads the quantized coefficient of the pixel of interest from the coefficient frame buffer 21, multiplexes it with the output of the selector 36, and outputs it. There is.

Therefore, in step S7, when the value stored in the status register 24 is 0, the valid data selection signal becomes 1 (when the pixel of interest is a feature point).
Includes the chain encoding circuit 12 (multiplexer 37)
Outputs chain coded data in which the coordinates (start point coordinates) of the pixel of interest as the feature point and the quantized coefficient at the feature point are multiplexed.

When the value stored in the state register 24 is 0 (hereinafter, appropriately referred to as the state 0), the fact that the valid data selection signal becomes 1 is a characteristic feature of the chain. Then, when the starting point was found. Then, in this case, as described above, the stored value of the state register 24 is set to 1 in step S7.

On the other hand, if it is determined in step S2 that the storage value of the status register 24 is not 0, the process proceeds to step S8, and it is determined whether or not the storage value is 1. If it is determined in step S8 that the value stored in the status register 24 is 1, the process proceeds to step S11 in FIG. 5, where the feature point data stored at the address generated by the address generator 32 is the mask frame. It is read from the buffer 22.

Here, the address generator 32 is the adder 3
An address having a value output from 8 or 39 as an upper address or a lower address is output. The adder 38 has a search X coordinate register 2
X coordinate stored in 7 and ROM (Search X-ROM) 3
An output value of 0 is supplied, and both are added and output there. Further, the Y coordinate stored in the search Y coordinate register 28 and the output value of the ROM (Search Y-ROM) 31 are supplied to the adder 39. As in the case, both are also added and output.

The ROMs 30 and 31 have a 3-bit address space in which the values shown in FIG. 6 are stored. The ROMs 30 and 31 use the output value of the 3-bit counter 29 as an address and output the stored value stored at that address.

Therefore, the outputs of the adders 38 and 39 are
(Output of adder 38, output of adder 39), and the value stored in search X coordinate register 27 or search Y coordinate register 28 is expressed as x or y, respectively, a 3-bit counter. When the output value of 29 is 0 to 7, the outputs of the adders 38 and 39 are
(X-1, y-1), (x-1, y), (x-, respectively)
1, y + 1), (x, y-1), (x, y + 1), (x
+1, y-1), (x + 1, y), (x + 1, y + 1)
Becomes

Therefore, as the output value of the 3-bit counter 29 changes from 0 to 7, the address generator 32 causes the search X coordinate register 27 and the search Y to be changed, as shown in FIG.
Eight pixels P0 adjacent to the pixel of the coordinate represented by the stored value of the coordinate register 28 (the hatched portion in the drawing)
The address corresponding to the coordinates of P7 to P7 is output.

From the above, in step S11, the pixel adjacent to the pixel of the coordinates represented by the stored values of the search X coordinate register 27 and the search Y coordinate register 28 (the pixel corresponding to the count value output by the 3-bit counter 29). The feature point data in any of P0 to P7 will be read from the mask frame buffer 22. here,
The process described below is performed by the search X coordinate register 27 and the search Y.
Based on the pixel of the coordinate represented by the value stored in the coordinate register 28, the pixel adjacent to the pixel is searched (sequentially as a target pixel), so the search X coordinate register 27
The pixel at the coordinate represented by the stored value of the search Y coordinate register 28 will be hereinafter referred to as a search center pixel, as appropriate.

When the characteristic point data is read out, the process proceeds to step S12, and it is determined whether or not the value is 1.
If it is determined in step S12 that the feature point data is not 1, that is, if the pixel of interest is not a feature point, the process proceeds to step S13, and it is determined whether the count value (output value) of the 3-bit counter 29 is 7. To be done. If it is determined in step S13 that the count value of the 3-bit counter 29 is not 7, that is, if processing has not been performed for all pixels adjacent to the search center pixel, the process proceeds to step S14. Is incremented by 1 and the process returns to step S2 in FIG. In this case, the value stored in the status register 24 is 1
As it is, the processing from step S11 onward is performed again through step S2 and step S8.

When it is determined in step S13 that the count value of the 3-bit counter 29 is 7, that is, all the pixels adjacent to the search center pixel are processed, and as a result, the pixels adjacent to the search center pixel are processed. When it is determined that the pixel has no feature point, the process proceeds to step S15, 0 is set in the state register 24, and step S2 is performed.
Return to

On the other hand, if it is determined in step S12 that the feature point data is 1, that is, if the target pixel is a feature point, the process proceeds to step S16, and the target pixel is newly set as the search center pixel. It That is, the stored value of the search X coordinate register 27 or the search Y coordinate register 28 is
Each is updated to the x or y coordinate of the pixel of interest. Then, in step S17, the valid data selection signal is set to 1 and the value stored in the status register 24 is set to 2. Further, in step S17, the 3-bit counter 29 is initialized to 0, and the feature point data having a value of 1 and stored in the mask frame buffer 22 corresponding to the pixel of interest is rewritten to 0. Return to S2.

Here, when the selector 36 receives the valid data selection signal having a value of 1, it refers to the value stored in the state register 24, and if it is 1, the starting point coordinate from the direction searcher 23. , The output of the ROM 33 from the outputs of the ROM 33 or the output of the direction change signal generator 35, and outputs the selected output to the multiplexer 37.

The ROM 33 has a 3-bit address space, like the above-mentioned ROMs 30 and 31, in which values as shown in FIG. 8 are stored. And R
The OM33, like the ROMs 30 and 31, again
The output value of the 3-bit counter 29 is used as an address, and the stored value stored at that address (hereinafter, this stored value is referred to as direction data) is output.

Therefore, when the output value of the 3-bit counter 29 is 0 to 7, the direction data C is read from the ROM 33.
0 to C7 are output respectively. This direction data C0
C7 to C7 represent the upper left, left, lower left, upper, lower, upper right, right, or lower right directions centered on the search center pixel. That is, the direction data C0 to C7 represent the directions in which the pixels P0 to P7 adjacent to the search center pixel are located, as shown in FIG. 9, for example.

Therefore, the direction data representing the direction of the pixel of interest with respect to the search center pixel is output from the ROM 33. In addition to the selector 36, the direction data is
The latch circuit 34 and the direction change signal generator 35 are also supplied.

From the above, in step S17, when the value stored in the state register 24 is 1, and the valid data selection signal becomes 1 (when the pixel of interest is a feature point), the chain encoding circuit 12 (Multiplexer 3
From 7), direction data indicating the direction in which the pixel of interest adjacent to the search center pixel (pixel located at the coordinates of the start point) and serving as the feature point is located, and the quantization coefficient of the pixel of interest. Is output as chain encoded data.

When the value stored in the state register 24 is 1 (hereinafter, appropriately referred to as state 1), the fact that the valid data selection signal becomes 1 is a characteristic feature of the chain. Then, the one adjacent to the starting point is found. In this case, as described above, the value stored in the status register 24 is set to 2 in step S17.

Returning to FIG. 4, if it is determined in step S8 that the value stored in the state register 24 is not 1, the process proceeds to step S21 in FIG. Where status register 2
An integer from 0 to 2 is stored in 4. Therefore, the value stored in the status register 24 is
It is determined in step S2 in FIG.
If it is determined in step S8 that the value is not 1, it means that the value is 2.

In step S21 of FIG. 10, as in the case of step S11 of FIG. 5, feature point data of any pixel (pixel corresponding to the count value output by the 3-bit counter 29) adjacent to the search center pixel. But,
It is read from the mask frame buffer 22, and the process proceeds to step S22, where it is determined whether or not the value is 1.
If it is determined in step S22 that the feature point data is not 1, that is, if the pixel of interest is not the feature point, the process proceeds to step S23, and it is determined whether the count value of the 3-bit counter 29 is 7. If it is determined in step S23 that the count value of the 3-bit counter 29 is not 7, that is, if processing is not performed for all pixels adjacent to the search center pixel, step S2
4, the count value of the 3-bit counter 29 is incremented by 1, and the process returns to step S2 in FIG. In this case, the value stored in the status register 24 remains 2, so that the values stored in step S2 and step S8 are
The processes in and after step S21 will be performed again.

If it is determined in step S23 that the count value of the 3-bit counter 29 is 7, that is, all the pixels adjacent to the search center pixel are processed, and as a result, the search center pixel is adjacent to the search center pixel. When it is determined that the pixel has no feature point, the process proceeds to step S25, 0 is set in the state register 24, and step S2
Return to

On the other hand, if it is determined in step S22 that the feature point data is 1, that is, if the pixel of interest is a feature point, the process proceeds to step S26 and the search X coordinate register 27 or the search Y coordinate register 28 is searched. The stored value is
The target pixel is newly set as the search center pixel by updating to the x or y coordinate of the target pixel. Then, the process proceeds to step S27, and the valid data selection signal is set to 1. Further, in step S27, the 3-bit counter 29 is initialized to 0, and the feature point data, which has a value of 1 and is stored in the mask frame buffer 22, is rewritten to 0. Return to S2. In this case, the status register 24
Since the stored value of 2 is still 2, the processing from step S21 onward is performed again through steps S2 and S8. However, in this case, step S26
At, the pixel of interest that was the target of the previous processing is
Since the pixel is newly set as the search center pixel, the pixel adjacent to the new search center pixel is processed as the pixel of interest.

Here, when the selector 36 receives the valid data selection signal having a value of 1, it refers to the value stored in the state register 24, and if it is 2, the start point coordinate from the direction searcher 23. , The output of the direction change signal generator 35 from the output of the ROM 33 or the output of the direction change signal generator 35, and outputs the selected output to the multiplexer 37.

The direction change signal generator 35 includes a ROM 33.
And the output of the latch circuit 34 are supplied there, and based on these inputs, direction change data, which will be described later, is generated.

That is, as shown in FIG. 11 (A), when the valid data selection signal becomes 1, the latch circuit 34 causes the ROM
The direction data (FIG. 11 (B)) output by 33 is latched, delayed by one clock, and then the direction data is changed to the direction change signal generator 35 until the valid data selection signal becomes 1. To continue to output (FIG. 11 (C)). That is, when a feature point is found (when the effective data selection signal becomes 1), the latch circuit 34 outputs direction data indicating the direction from the search center pixel to the feature point until the next feature point is found. Continue to be done. Here, the latch circuit 34
Hereinafter, the direction data output by will be referred to as forward direction data as appropriate.

When the effective data selection signal becomes 1, that is, when the characteristic point is found, the direction change signal generator 35 outputs the direction indicated by the forward direction data output from the latch circuit 34 and R.
The direction indicated by the direction data output from the OM 33 is compared, and the direction change data indicating the change in the direction indicated by the direction data with respect to the direction indicated by the front direction data is output in accordance with the comparison result. .

That is, the direction change signal generator 35 outputs D0 as the direction change data when the direction indicated by the forward direction data and the direction indicated by the direction data are the same as shown in FIG. 12A, for example. To do. In addition, the direction change signal generator 35 is, for example, as shown in FIGS.
When the direction indicated by the direction data differs from the direction indicated by the forward direction data by 45 degrees, 90 degrees, or 135 degrees counterclockwise, D1 to D3 are respectively output as the direction change data. Further, the direction change signal generator 35 has
As shown in (E) to (G), the direction indicated by the direction data is 90 degrees clockwise from the direction indicated by the forward direction data by 90 degrees.
In the case of a difference of 135 degrees or 135 degrees, D4 to D6 are output as the direction change data.

As for the relationship between the direction indicated by the direction data and the direction indicated by the forward direction data, in addition to the case described above, FIG.
As shown in FIG. 2 (H), the two directions may be different by 180 degrees. However, in the chain encoding circuit 12 of FIG. 2, once the feature point becomes a processing target, as described above, the corresponding feature Since the point data is rewritten to 0, the direction indicated by the direction data and the direction indicated by the forward direction data are 180
There will be no occasions where they differ. Therefore, as shown in FIG. 12H, when the direction indicated by the direction data and the direction indicated by the forward direction data are different by 180 degrees, no code is given as the direction change data (it is necessary to give the code). Absent).

In reality, the direction change signal generator 35 stores a correspondence table of the direction change data and the forward direction data and direction data as shown in FIGS. 13 and 14, and the latch circuit 34 stores the correspondence table. The forward direction data being output,
A line matching the combination with the direction data output from the ROM 33 is searched from the correspondence table, and the direction change data described in the right column of the line is output. Note that the direction change data D0 to D6 are assigned with code words as shown in FIG. 15, for example, and the code words are actually output.

From the above, in step S27, when the value stored in the status register 24 is 2 and the valid data selection signal becomes 1 (when the pixel of interest is a feature point), the chain encoding circuit 12 (Multiplexer 3
From 7), the direction about the feature point found last time (the direction from the feature point found two times before to the feature point found last time) and the direction about the feature point found this time (from the feature point found last time, found this time) Direction change data indicating the difference between the characteristic point and the target point), and chain encoded data in which the quantized coefficient at the characteristic point (pixel of interest) found this time is multiplexed.

When the value stored in the state register 24 is 2 (hereinafter, appropriately referred to as the state 2), the valid data selection signal becomes 1 when the number of consecutive characteristic points is 3 or more. This is the case when a feature point that constitutes a chain that is other than the start point and ones adjacent thereto (third feature point and subsequent ones) is found. In this case, the value stored in the status register 24 remains 2.

Returning to FIG. 4, if it is determined in step S3 that there are no unprocessed coordinates, the process ends. Then, after waiting for the quantized coefficient or feature point data for the next frame to be stored in the coefficient frame buffer 21 or the mask frame buffer 22, the processing from step S1 is performed again.

Next, the output of the selector 36 will be further described with reference to FIG. Now, for example, as shown in FIG. 16, when there is a chain formed by connecting five consecutive characteristic points (hatched portions in the figure),
Looking at the line scan order, the feature point located at the coordinate (i, j) first appears. Since the state when the feature point is first found is state 0, the selector 36 outputs the coordinate (i, j) of the feature point, which is output by the direction searcher 23, as the start point coordinate. After that, the state is set to the state 1, and the feature point (i + 1, j) adjacent to the feature point located at the coordinate (i, j) (hereinafter, appropriately described as the feature point (i, j)) is detected. . This feature point (i
+1, j) is located to the right of the previously found feature point (i, j), and the state is state 1. Therefore, the selector 36 outputs the right direction data C6 output from the ROM 33. (FIG. 9) is output. And in this case,
The state is set to state 2.

Next, the feature point (i + 2, j + 1) adjacent to the feature point (i + 1, j) is detected. In this case, since the state is the state 2, the direction change data indicating the difference between the direction of the previously found feature point and the direction of the currently found feature point is output from the selector 36.
That is, the direction of the feature point (i + 1, j) found last time is the right direction represented by the direction data C6 as described above, and the feature point (i + 2, j +) found this time.
The direction of 1) is the characteristic point (i + 2, j + 1)
Is located in the lower right direction of the feature point (i + 1, j),
It is represented by the direction data C7. Therefore, the forward direction data or the direction data in the correspondence table shown in FIG. 14 is C6, respectively.
Alternatively, the direction change data D4 described in the right column of the row of C7 is output from the selector 36.

Further, the feature point (i + 2, j + 1) is
Since the feature points (i + 3, j + 2) are adjacent to each other, this feature point (i + 3, j + 2) is detected. The direction in which the feature point (i + 3, j + 2) is located is the direction data C for the previously found feature point (i + 2, j + 1).
It is the same lower right direction as the direction indicated by 7, and therefore the direction for the feature point (i + 3, j + 2) found this time is
It is represented by the direction data C7. In addition, since the state remains the state 2, in this case, from the selector 36, the forward direction data and the direction data in the correspondence table shown in FIG. Direction change data D0 being output.

Then, for the feature points (i + 2, j + 3) adjacent to the feature point (i + 3, j + 2), the feature point (i
+3, j + 2), and as a result, the selector 36 outputs direction change data D6.

Therefore, when there is a chain on the image, the chain encoding circuit 12 first multiplexes the coordinates of the feature point which is the starting point of the chain and the quantized coefficient at the feature point. Stuff is output. Then, the direction data of the feature point adjacent to the start point and the quantized coefficient at the feature point are multiplexed and output. If there is a feature point (excluding the start point) adjacent to the feature point, that is, if there is a third feature point or later, the direction change data about the feature point and the feature point And the quantized coefficient in 1 is multiplexed and sequentially output.

As described above, the chain encoding circuit 1
As described above, the chain-encoded data output from No. 2 is supplied to the RAMs 14 _{1 to} 14 _N via the selector 13 (FIG. 1), so that the RAM 14 _n receives the n-th frame of the chain-encoded data. The data is stored. And
When the chain coded data is stored in all of the RAMs 14 _{1 to} 14 _N , that is, when the chain coded data for N frames is stored, the chain replacement circuit 15 reads the chain coded data, and a process as described in detail below. Is applied.

FIG. 17 shows a configuration example of the chain replacement circuit 15 of FIG. The chain map circuit 121 is R
When N frames of chain-encoded data are stored in the AM 14 _{1 to} 14 _N , the chain-encoded data of the first frame is read from the RAM 14 ₁ and the RAM (Chain RA
M) 122 for storage.

Furthermore, the chain map circuit 121, for the chain encoded data of the second to Nth frames, the start point coordinates, the direction data, and the direction change data included therein (hereinafter, these are collectively summarized as appropriate, (Referred to as a direction component), a bitmap of a chain connecting consecutive feature points existing in the second to Nth frames (hence, the chain is a continuous point (continuous point sequence) in the image) , RAM (MapRAM) 123 _{1 to} 123 _N-1 respectively (the bitmap of the nth frame is
Deploy on AM123 _n-1 (however, in this case, n is 2
No less than N and no more than N)).

That is, for example, as shown in FIG. 18, the chain map circuit 121 stores _{1 in} the address of the RAM 123 _n-1 corresponding to the position where the chain of the nth frame exists, and the position where the chain does not exist. 0 is stored in the address corresponding to.

The chain encoded data of the first frame is stored in the RAM 122, and the RAMs 123 _{1 to} 123 _N-1.
Then, when the bitmaps for the chains of the 2nd to Nth frames are respectively developed, in the replacement circuit 124,
The process according to the flowchart of FIG. 19 is performed for each chain existing in the first frame stored in the RAM 122.

That is, first, in step S121, the RAM (motion parameter RAM) 126 is initialized, the process proceeds to step S122, the first frame (head frame) is set as the frame of interest, and the process proceeds to step S123. In step S123, the frame of interest, that is, the first frame
A motion parameter existing in a frame from a chain (chain of interest) to a subsequent frame (frame next to the frame of interest) is calculated. In the calculation of the motion parameter in step S123, the replacement circuit 124 stores the data required for the calculation of the motion parameter in the RAM 12
2, 123 ( _{one of the} RAMs 123 _{1 to} 123 _N ) is read out and output to the motion parameter calculation circuit 125 so that the motion parameter calculation circuit 125 can perform the operation. The details of the motion parameter calculation circuit 125 will be described later with reference to FIGS. 29 to 36.

After calculating the motion parameters, step S12
In step 4, the importance (visual importance) of the chain of interest in the subsequent frame is calculated. That is, in step S124, as shown in FIG. 47A, for example, the chain of interest is moved (converted) according to the motion parameter obtained in step S123, and in the subsequent frame, as shown in FIG. , The range of L1 × L2 pixels (the range surrounded by the dotted line in the figure) centered on each pixel (the hatched part in the figure) forming the target chain (moving chain) after the movement The number of existing feature points is calculated. Further, in step S124, after the number of characteristic points existing in the range of L1 × L2 pixels centered on each pixel forming the moving chain in the subsequent frame is calculated, the average value thereof (each of the forming points forming the moving chain is calculated). The sum of the number of feature points in the subsequent frame existing in the range of L1 × L2 pixels centered on the pixel divided by the number of pixels forming the moving chain) is obtained, and this is taken as the importance.

Here, in the RAM 123 _n , as shown in FIG.
As shown in FIG. 8, 1 is stored in the addresses corresponding to the pixels (feature points) forming the chain existing in the (n + 1) th frame, and 0 is stored in the other addresses. The nth frame
If it is a frame, in step S124, the RAM 12
The bit map stored in 3 _n is referred to, and L1 × centered on each pixel forming the moving chain among them.
By calculating the sum of the stored values of the addresses corresponding to the range of L2 pixels (for example, L1 = L2 = 3), in the subsequent frame (in this case, the (n + 1) th frame), each pixel forming the moving chain is centered. The number of feature points existing in the range of L1 × L2 pixels is calculated.

The importance calculated as described above can be said to be the density of feature points existing in the range of L1 × L2 pixels centering on each pixel forming the moving chain in the subsequent frame. As the degree of importance, it is possible to adopt an amount other than the density of such feature points.

That is, for example, in the subsequent frame, of the pixels that make up the moving chain,
It is possible to divide the number of feature points existing in the range of × L2 pixels by the number of pixels forming the moving chain, and use the value obtained as a result as the importance. Also, for example,
The second stored in RAM 20 _{1 to} 20 _N-1 (FIG. 1)
Through reference to the image data of the Nth frame, the edge strength of each pixel corresponding to each pixel forming the moving chain in the subsequent frame (the edge strength of each pixel existing in the moving chain in the following frame) is calculated, and It is also possible to use the average value of the edge strength as the importance.

After the importance is calculated, the process proceeds to step S125, and it is determined whether the importance is equal to or higher than a predetermined threshold value T. When it is determined in step S125 that the degree of importance is not equal to or higher than the predetermined threshold T, the subsequent frame is
Assuming that there is no chain that constitutes the contour of the same object as the target chain (a chain that (substantially) overlaps the target chain after movement), steps S126 to S128 are skipped, and the process proceeds to step S129. Step S1
25, if it is determined that the degree of importance is equal to or higher than the predetermined threshold value T, that is, if there is a chain that (substantially) overlaps the target chain after the movement in the rear frame, the process proceeds to step S126, and the chain of the frame thereafter. Is replaced by the moving chain (this replacement means that the chain in the rear frame is replaced by the moving chain and is recognized in the replacing circuit 124), whereby the target chain and the rear chain are recognized. It is associated with a chain of frames.

Further, in step S126, the motion parameter for the subsequent frame calculated in step S123 is stored in the RAM 126, and the process proceeds to step S127. In step S126, the RAM 123 is also updated. That is, if the chain existing in the rear frame is replaced by the moving chain (since the moving chain is used instead of the chain),
It is not necessary to process that chain. Therefore, in step S126, for example, assuming that the current frame is the nth frame, L1 × L2 centered on each pixel forming the moving chain in the bitmap of the (n + 1) th frame stored in the RAM 123 _n. The stored values of the addresses corresponding to the pixel range are all set to 0. As a result, in the (n + 1) th frame, the characteristic points in the range of L1 × L2 pixels centered on each pixel forming the moving chain are erased, so to speak.

Then, in step S127, the subsequent frame is newly set as the frame of interest, and step S128.
To the Nth frame (final frame).
Is determined. If it is determined in step S128 that the frame of interest is not the Nth frame, the process returns to step S123, and the frame of interest is newly included in step S127.
The process after step S123 is performed using the chain (moving chain) replaced in 126 as the target chain.

If it is determined in step S128 that the new frame of interest is the Nth frame,
Proceeding to step S129, the chain number of the chain which is the frame of interest (hence, the first frame) which is the frame of interest in step S122, which is the chain of interest first, and the motion parameter stored in the RAM 126 (to the next frame) Motion parameter) and the number of motion parameters are output to the select multiplexing circuit 16, and the process ends.

As described above, the processing shown in FIG. 19 is performed for all chains existing in the first frame.

In the select multiplex circuit 16, as described above, the chain encoded data of the chain number output from the chain permutation circuit 15 (substitution circuit 124) is RA.
The chain encoded data read from M14 ₁ is multiplexed with the motion parameter and the number thereof which are also output from the chain permutation circuit 15 to be output as multiplexed chain encoded data.

[0110] That is, for example, now, as shown in FIG. 21, when the chain CH ₁ to CH _N respectively present in the first through the N-th frame are associated, from the select multiplexing circuit 16, the chain CH ₁ Chain-encoded data about, and chain CH ₁ to chain CH _N-1
Parameters to the subsequent frame and the number N−
A multiplexed version of 1 and 1 is output.

Therefore, the chain that can be replaced (correlated) with the chain of interest of the first frame is the coded data for the chain of interest, and the chain is moved (motion compensated) to restore another chain. Since it is output as the motion parameter of, it is possible to reduce the redundancy in the time direction as compared with the case of outputting the chains existing in each frame as chain encoded data as in the conventional case. As a result, the compression efficiency of the image can be improved.

The RAM 122 shown in FIG.
As in the case of 123 _{1 to} 123 _N-1 , it is possible to expand the bitmap of the first frame. However, in this case, the RAM 122 needs to be a frame memory capable of storing a bit map for one frame, and therefore needs to have a large storage capacity. On the other hand, as described above, the RAM1
In the case where the chain encoded data existing in the first frame is stored in 22, the chain encoded data for one frame usually has a smaller data amount than the bitmap for one frame. , RAM122
Also, it is possible to use a memory having a small storage capacity.

By the way, according to the image coding apparatus of FIG. 1, it is possible to improve the compression efficiency of image data as described above, but only the chains existing in the first frame are shown in FIG. Because the processing is performed, the second
When a chain first appears in a frame subsequent to the frame (for example, when the scene changes in the second frame and subsequent frames), the information about the chain is not output from the select multiplexing circuit 16, and accordingly, ,
It becomes difficult for the decryption side to decrypt the chain.

Therefore, FIG. 22 shows the configuration of the second embodiment of the image coding apparatus to which the present invention is applied. In the figure, the same reference numerals are given to portions corresponding to the case in FIG. That is, this image encoding device is the same as the image encoding device of FIG. 43 except that a chain replacement circuit 215 or a select multiplex circuit 216 is provided instead of the chain replacement circuit 15 or the select multiplex circuit 16. It is similarly configured.

The chain replacement circuit 215 performs the same processing as the chain replacement circuit 15 of FIG. 1 on not only the chains existing in the first frame but also the chains existing in the second and subsequent frames. As will be described later, the frame number, chain number, motion parameter, and the number of motion parameters are output to the select multiplexing circuit 216.

The select multiplex circuit 216 reads out the chain-encoded data of the chain corresponding to the chain number from the chain permutation circuit 215 from the RAM 14 _n similarly having the frame number n from the chain permutation circuit 215 as a suffix. The chain encoded data, the motion parameter from the chain replacement circuit 215 and the number thereof are multiplexed to form multiplexed chain encoded data, which is output to the buffer 17.

FIG. 23 shows the chain replacement circuit 21 of FIG.
5 shows a configuration example of No. 5. In the figure, parts corresponding to those in the chain replacement circuit 15 of FIG. 17 are designated by the same reference numerals. That is, the chain replacement circuit 215 is replaced by the replacement circuit 13 instead of the replacement circuit 124.
2 is provided, and further, a new chain encoding circuit 131
_{Our 1} to 131 _N-1 and the chain decoding circuit 133 ₁ through 133 _N-1 are newly provided, has the same configuration as the chain substitution circuit 15 of FIG. 17.

In the chain replacement circuit 215 configured as described above, the RAM is replaced by the RAM in the same manner as in FIG.
To 122, the chain coded data of the first frame is stored, in the RAM 123 ₁ through 123 _N-1, the bit map for the second to the chain of the N-th frame is deployed, respectively, in a substitution circuit 132, FIG. 24 The process according to the flowchart of FIG.

That is, first, in step S131, a variable n for counting the number of frames is set to 1 as an initial value, and the process proceeds to step S132 to perform a replacement process based on the nth frame as described later. Be seen. Then, when the replacement process based on the nth frame is completed, the variable n is incremented by 1 in step S133, the process proceeds to step S134, and the variable n
Is less than or equal to N, which is the number of frames in a processing unit. In step S134, the variable n is N
When it is determined that the following is true, the process returns to step S132, and the replacement process based on the nth frame is performed again.
If it is determined in step S134 that the variable n is not equal to or less than N, the process ends.

Next, FIG. 25 is a flow chart showing details of the replacement process based on the nth frame of FIG. In the replacement process based on the nth frame, step S14
In 1 to S148, the same processes as in steps S121 to S128 of FIG. 19 are performed, respectively, except that the nth frame is set as the frame of interest instead of the first frame in step S142.

Here, since the chain coded data for the chain existing in the first frame is stored in the RAM 122, the chain coded data stored in the RAM 122 is used for the chain existing in the first frame. , It is possible to recognize each chain existing in the first frame and calculate the motion parameter from that chain to a subsequent frame in step S143. bitmap with 1 value stored at the address corresponding to the positions of feature points that make up the can, since only stored in the RAM 123 ₁ through 123 _N-1, which recognizes the chains present in the second frame and subsequent frames In the replacement circuit 132, the bit Flops scans the stored contents of RAM123 being deployed, it is necessary to reconstruct the portion 1 is continuous as a chain, it takes time for processing.

[0122] Further, the bit map stored in RAM 123 ₁ through 123 _N-1, in step S146, but is updated as in step S126 described in FIG. 19, the substitution circuit 132, after the update Need to recognize the chain of.

Therefore, in step S142, the second
When the frame after the frame is set as the frame of interest,
The replacement circuit 132 recognizes a chain existing in the frame of interest as follows.

That is, for example, if it is assumed that the n-th frame is the frame of interest in step S142 (here, n is an integer of 2 or more and N or less),
The control signal is output to the new chain encoding circuit 131 _n-1 configured similarly to the chain encoding circuit 12 described in FIG. The chain encoding circuit 131 _n-1 is the replacement circuit 1
When the control signal is received from 32, the chain decoding circuit 1
33 _n-1 is controlled, whereby the chain encoded data of the nth frame is read from the RAM 14 _n and the chain is decoded. The image data of the n-th frame obtained by being chain-decoded by the chain decoding circuit 133 _n-1 is supplied to the new chain encoding circuit 131 _n-1 , where the contents stored in the RAM 123 _n-1 are stored. The chain is coded again with reference.

That is, from the chain decoding circuit 133 _n-1 , the nth frame image data and the nth frame chain encoded data obtained by decoding are decoded.
The feature point of the frame is also the new chain encoding circuit 131.
_In the new chain encoding circuit 131 _n-1 , the chain decoding circuit 133 _n-1 is supplied to the _n-1.
The point (pixel) which is the feature point of the n-th frame from No. ₁ and whose stored value of the address of the RAM 123 _n-1 corresponding to the feature point is 1 is, so to speak, a true feature point, and is referred to as a chain decoding circuit. The image data of the n-th frame supplied from 133 _n-1 is chain-coded, and the chain-coded data obtained as a result is output to the replacement circuit 132.

Therefore, the new chain encoding circuit 131
_{The n-} th frame chain encoded data output from _n-1 reflects the updated contents of the RAM 131 _n-1 .

In the replacement circuit 132, as described above,
R supplied from the new chain encoding circuit 131 _n-1
The chain existing in the n-th frame is recognized based on the chain-encoded data of the n-th frame that reflects the updated contents of AM131 _n-1 .

Then, in the replacement circuit 132, step S
In 149, as in the case of step S129 of FIG. 19, the chain number of the chain initially set as the target chain, the motion parameter (motion parameter to the subsequent frame) stored in the RAM 126, and the number of motion parameters are The frame number of the frame in which the chain first set as the target chain exists (the frame set as the target frame in step S142) is output to the select multiplex circuit 216, and is also output to the select multiplex circuit 216.

The process shown in FIG. 25 is performed in step S
This is performed for all chains existing in the nth frame which is the frame of interest at 142.

In the select multiplex circuit 216, as described above, the RAM 14 _n having the frame number n output from the chain replacement circuit 215 as a suffix, and the chain corresponding to the chain number also output from the chain replacement circuit 215 Of the chain encoded data is read out, and the chain encoded data is multiplexed with the motion parameter and the number thereof from the chain replacement circuit 215 to be output as multiplexed chain encoded data.

That is, for example, as shown in FIG. 26, assuming that the chains CH _{1 to} CH _N existing in the _first to Nth frames are associated with each other, from the select multiplexing circuit 216 in FIG. As in the case
The chain coded data for the chain CH _1, obtained by multiplexing the motion parameters and the number N-1 thereof to the frame after the chain CH ₁ to the chain CH _N-1 is output. Furthermore, as shown in FIG. 2, even if the chain CH ₂ 'has appeared for the first time in the second frame,
By performing the “replacement process based on the second frame” in step S132 of 4, the chain CH ₂ '
Is associated with the chain to be associated with it (in the embodiment of FIG. 26, the chain CH ₃ ′ of the third frame), and as a result, the select multiplex circuit 216 outputs the chain code for the chain CH ₂ ′. The multiplexed data, the motion parameter for the subsequent frame of the chain CH ₂ 'and its number 1 are also output.

Therefore, in this case, even if the chain appears for the first time in the second and subsequent frames, the decoding side can decode the chain.

A decoded image can be obtained from the multiplexed chain coded data output from the image coding apparatus shown in FIGS. 1 and 22 as follows. That is, the chain encoded data and the motion parameter are extracted from the multiplexed chain encoded data, and the chain encoded data is chain decoded. Then, the decoded image (chain) obtained as a result is motion-compensated according to the motion parameter. As a result, the decoded image of the frame following the first decoded image can be obtained.

However, the decoded image is as shown in FIG. 27 and FIG.
As shown in FIG. That is, if a chain output as chain encoded data from the image encoding device is called a basic chain, for example, as shown in FIG.
As shown in (A), in the original image, when a certain chain is present in the nth frame and three chains which are obtained by dividing the chain are present in the rear frame, the nth frame The chain is the basic chain,
When associated (replaced) with the three chains of the subsequent frame, the decoding side decodes the chains of the subsequent frame by performing motion compensation on the basic chain.
The three chains of the subsequent frame are decoded as one chain as shown in FIG.

Further, for example, as shown in FIG. 28A, in the original image, there are three chains in the nth frame, and in the rear frame, one chain is formed by connecting the chains. If present, each of the three chains in the nth frame is a base chain,
When associated with (replaced by) one chain of the subsequent frame, the decoding side decodes the chain of the subsequent frame by motion-compensating the basic chain. As shown in the figure (B), it is decoded as three chains.

Next, FIG. 29 shows a configuration example of the motion parameter calculation circuit 125 of FIG. 17 (FIG. 23). The motion parameter calculation circuit 125 is used in the confidence point calculation circuit 41.
And a motion parameter calculation circuit 42. The confidence point calculation circuit 41 is supplied from the replacement circuit 124 (or 132) with the chain-coded data of the chain for which the motion parameter is to be obtained, and the RAM 2
Image data is supplied from 0. Then, when the encoded code data is input, the confidence point calculation circuit 41 detects a chain corresponding to the chain encoded data, and calculates the reliability representing the reliability of each of the feature points forming the chain (continuous points). Based on the reliability, two points are selected from the feature points in the order of high reliability (hereinafter, the highest reliability point or 2 points out of the two selected points). The second highest point is the first
Confidence point or second confidence point). Further, the confidence point calculation circuit 41 detects the motion vector of each of the first and second confidence points, and together with the coordinates of the first and second confidence points,
It is adapted to be output to the motion parameter calculation circuit 42.

Here, in the present specification, the motion vector means a vector representing the amount of parallel movement of an object (to be exact, the amount of parallel movement and the moving direction) when the object is translated. On the other hand, the motion parameter is not only the amount of translation when the object is translated, that is, the motion vector, but also the scaling factor (enlarging or reducing rate) when the object is scaled up or down, and when it is rotated. It also means a parameter indicating a rotation amount (more precisely, a rotation amount and a rotation direction (however, if the rotation direction is set to one direction, the rotation direction is not necessary)).

The motion parameter calculation circuit 42 receives from the confidence point calculation circuit 41 the first confidence point (first
When the coordinate of the trust point) and its motion vector and the second confidence point (coordinate of the first confidence point) and its motion vector are received, the parallel movement amount, the scaling ratio, and the rotation amount of the chain are calculated based on them. A motion parameter to be expressed is calculated.

FIG. 30 shows a configuration example of the confidence point calculation circuit 41 of FIG. Chain encoded data is input to the shape reliability calculation circuit 51,
The shape reliability calculation circuit 51 is configured to calculate the reliability of the feature point of the chain corresponding to the chain encoded data, based on the geometrical shape around each feature point forming the chain. ing. Furthermore, the shape reliability calculation circuit 51 is a point that is a candidate for a first or second reliability point (hereinafter, simply referred to as a reliability point) from the characteristic points forming the chain based on the obtained reliability ( Hereinafter, the reliability candidate point will be appropriately detected) and output to the prediction error reliability calculation circuit 52.

Specifically, the shape reliability calculation circuit 51
In order to distinguish the angle formed by the chain with each feature point forming the chain as the apex from the reliability (hereinafter, appropriately, other reliability (prediction error reliability, which will be described later in this embodiment)), the angle reliability is calculated. , And the angle reliability (in the present embodiment, the angle itself, as described above) takes a value within a predetermined range as the reliability candidate point.

The prediction error reliability calculation circuit 52 calculates the motion vector of the reliability candidate point from the shape reliability calculation circuit 51 by performing so-called block matching with reference to the current frame buffer 54 and the subsequent frame buffer 55. Then, based on the prediction error obtained at that time, in order to distinguish the reliability of each reliability candidate point (hereinafter, appropriately, from other reliability (the above-described angle reliability in this embodiment), the prediction error reliability Is called) is calculated. In this embodiment, for example, the average value of the absolute values of prediction errors is used as the prediction error reliability.

Furthermore, the prediction error reliability calculation circuit 52
Based on the reliability of prediction error, the first
And the second confidence point are selected and output to the motion parameter calculation circuit 42 together with their motion vectors.

The read circuit 53 reads from the RAM 20 the image data (hereinafter, referred to as attention frame data) of a frame (attention frame) in which a chain (attention chain) currently being processed exists (hereinafter referred to as attention frame data) and a subsequent frame (attention). The image data (post frame data) of the frame next to the frame of interest is read out, and the frame data of interest or the post frame data is stored in the current frame buffer 54 or the post frame buffer 55, respectively.

Next, the operation will be described. When the chain encoded data of the chain for which the motion parameter is to be obtained is input from the replacement circuit 124 (or 132), the chain encoded data is supplied to the shape reliability calculation circuit 51. Further, at this time, the reading circuit 5
Reference numeral 3 indicates a frame in which a chain corresponding to the input chain-encoded data exists, that is, image data of a frame of interest (frame of interest data) and subsequent frame data.
It is read from the RAM 20 and supplied to the current frame buffer 54 or the subsequent frame buffer 55 to be stored therein.

Thereafter, the shape reliability calculation circuit 51 performs the processing according to the flowchart of FIG. That is, in the shape reliability calculation circuit 51, first, in step S31, the chain (target chain) corresponding to the chain encoded data is detected based on the direction component of the input chain encoded data. further,
In step S31, a number (hereinafter, appropriately referred to as a pixel number) is serially assigned to each pixel (characteristic point) forming the target chain, for example, from the start point to the end point. That is, for example, if the chain of interest is composed of P pixels, 1 is attached to the starting point, 2 is attached to the point next to the starting point, ..., And the p-th point from the starting point. Is attached with n, ..., P is attached to the end point.

Then, in step S32, it is determined whether or not there is an unprocessed pixel among the pixels forming the target chain. When it is determined in step S32 that there is an unprocessed pixel, when the pixels forming the target chain are viewed in ascending order of pixel number, for example, the first unprocessed pixel detected is set as the target pixel. Then, the process proceeds to step S33.

In step S33, of the pixels forming the chain of interest existing within the range of M1 × M2 pixels centered on the pixel of interest P _c , the pixel having the largest pixel number (hereinafter, appropriately referred to as the first reference). A point) P _max and a minimum pixel (hereinafter appropriately referred to as a second reference point) P _min are detected. That is, for example, as shown in FIG.
In the case where the chain composed of pixels with the image number of (the shaded portion in the drawing) is the target chain, when the pixel of pixel number 13 is the target pixel, Of the pixels forming the chain of interest, which exists within the range of M1 × M2 pixels centered on the pixel of interest (in the range indicated by the bold frame in the figure), the pixel with the maximum pixel number 16 is the first reference The pixel having the smallest pixel number 9 is detected as the point P _max and the second reference point P _min . In the example of FIG. 32, M1 = M2 = 7.

After that, the process proceeds to step S34, and the angle θ formed by the points P _c , P _max and P _min with the target pixel P _c as the apex.
(= ∠P _min P _c P _max ) is calculated as the angle reliability (here, the unit of θ is degree). That is, FIG.
When the target pixel P _c , the first reference point P _max , and the second reference point P _min as shown in FIG. 33 are obtained, the angle θ as shown in FIG. 33 is calculated as the angle reliability. . here,
In step S34, the angle (angle reliability) θ is specifically calculated as follows, for example. That is, the coordinates of the points P _c , P _max , and P _min are (x, y), respectively.
(X _a, y _a), when the (x _b, y _b), the angle reliability θ
Is calculated according to the formula θ = | arctan (y _a -y / (x _a -x))-arctan (y _b -y / (x _b -x)) |.

Further, in step S34, θ0 <θ1
(Here, the units of θ0 and θ1 are degrees), the angular reliability θ is greater than θ0 and less than θ1, or greater than 180−θ1 and 180.
It is determined whether it is less than −θ0. Step S34
, The angle reliability θ is larger than θ0, and θ1
Not less than or greater than 180-θ1 and 180
If it is determined that it is not less than −θ0, step S32.
Return to Further, in step S34, the angle reliability θ
Is determined to be greater than θ0 and less than θ1, or greater than 180−θ1 and 180−θ0.
If it is determined that it is less than,
The pixel of interest P _c (coordinates of the pixel of interest P _c ) that gives the angle reliability θ is output to the prediction error reliability calculation circuit 52 as a reliability candidate point.

Here, θ0 or θ1 is, for example, about 20 degrees or 70 degrees, respectively. This is for the following reason. That is, when the angular reliability θ corresponding to the pixel of interest P _c is around 45 degrees (for example, about 20 degrees to 70 degrees) or around 135 degrees (for example, about 110 degrees to 160 degrees), Target pixel P _c
It has been empirically known that the motion vector of is highly reliable when it is obtained by performing block matching, for example. On the other hand, in the prediction error reliability calculation circuit 52 in the subsequent stage, the motion vector is detected by block matching, as described later. Therefore, in order to detect a highly reliable motion vector in the prediction error reliability calculation circuit 52, θ0 or θ1 is set to about 20 degrees or 70 degrees, respectively, as described above. From this, it can be said that a pixel whose angle reliability θ is larger than θ0 and smaller than θ1 or larger than 180−θ1 and smaller than 180−θ0 has high geometrical reliability. You can

After the reliability candidate points are output in step S35 as described above, the process returns to step S32, and steps S32 to S35 are repeated until it is determined in step S32 that there are no unprocessed pixels. The process of is repeated. Then, in step S32, when it is determined that there is no unprocessed pixel, the process is ended, waits for new chain encoded data to be input, and
The processing from step S31 is repeated.

In step S33, if the first reference point and the second reference point are not obtained, the subsequent processing is performed, the process returns to step S32, and step S3 is performed again.
The processing from step 2 is repeated.

The prediction error reliability calculation circuit 52 waits for the reliability candidate points for one chain to be output from the shape reliability calculation circuit 51, and then performs the processing according to the flowchart of FIG. That is, when the prediction error reliability calculation circuit 52 receives all the reliability candidate points for one chain from the shape reliability calculation circuit 51, in step S41, the shape reliability calculation circuit 52.
It is determined whether or not all of the candidate trust points for a chain received from the unprocessed ones. In step S41, the shape reliability calculation circuit 52
If it is determined that there is an unprocessed reliability candidate point received from, the reliability candidate point (unprocessed reliability candidate point) is first detected when looking at the ascending order of pixel numbers, for example. The selected one is set as the focused trust candidate point, and step S4
Proceed to 2.

In step S42, a motion vector is detected by block matching for a range of N1 × N2 pixels centered on the target reliable candidate point. That is,
Assuming that the coordinates of the focused reliability candidate point are (x, y), the focused frame data in the range of N1 × N2 pixels centered on the focused reliability candidate point (x, y) is extracted from the current frame buffer 54 in step S42. At the same time as the reading, the rear frame data in the range of N1 × N2 pixels centering on the point (x + Δx, y + Δy) is read from the rear frame buffer 55. Further, in step S42, each of the target frame data in the range of N1 × N2 pixels centered on the point (x, y) and the range of N1 × N2 pixels centered on the point (x + Δx, y + Δy) are selected. The difference with each of the subsequent frame data, that is, the prediction error is calculated, and then, for example, the average value of the absolute values of the prediction errors is calculated.

Then, in step S42, the above processing is repeated while changing Δx or Δy within a predetermined range, and Δx within the predetermined range is repeated.
Or △ y for each obtains a mean value of the absolute values of the prediction error, gives the minimum value _{(△ x min, △ y min} )
Is calculated as the motion vector of the target reliability candidate point (x, y) to the subsequent frame.

After that, the process proceeds to step S43, and the reliability (prediction error reliability) of the target reliable candidate point is calculated based on the prediction error obtained when the block matching is performed in step S42. That is, in the present embodiment, in step S43, for example, the motion vector (Δx
_min , Δy _min ) is obtained, the average value of the absolute values of the prediction errors is directly used as the prediction error reliability. And
Returning to step S41, the processing from step S41 is repeated again.

As the prediction error reliability, it is possible to use, in addition to the average value of the absolute values of the prediction errors, for example, the sum of absolute values of the prediction errors or the sum of squares.

After that, in step S41, when it is determined that there is no unprocessed reliability candidate point received from the shape reliability calculation circuit 52, that is, for all the reliability candidate points of one chain, When the calculation of the motion vector and the prediction error reliability is completed, step S4
4, the prediction error reliability value is the smallest (the prediction error reliability is the highest (in the present embodiment, the prediction error reliability is
The smaller the value is, the higher the reliability is.)) The reliability candidate point and the second smallest value thereof (the prediction error reliability is 2).
The second highest reliability candidate point is the most reliable point among the points (feature points) forming the chain (the chain corresponding to the chain encoded data input to the shape reliability calculation circuit 51). The motion vector is selected and output as the first confidence point or the second confidence point, respectively, together with the motion vector.

When no confidence point is obtained, or only the first confidence point is obtained (when two or more confidence candidate points are not output from the shape reliability calculation circuit 51, that is, step In the case where two or more points satisfying the condition described in S34 do not exist), the start point or the end point of the chain is set as the first or second confidence point.

The first and second obtained as described above
The confidence points and their motion vectors are supplied to the motion parameter calculation circuit 42 (FIG. 29) as described above.

The motion parameter calculation circuit 42 calculates the first from the confidence point calculation circuit 41 (prediction error reliability calculation circuit 52).
Upon receiving the second confidence point and their motion vectors, the motion parameters for the subsequent frames of the chain are calculated according to the following equation.

That is, the coordinates of the first or second confidence point are (x1, y1) or (x2, y2), and the motion vectors of the first or second confidence point are (MVx1, MV).
y1) or (MVx2, MVy2), a movement parameter (parallel movement vector) indicating the parallel movement amount of the chain (MVx, MVy), a movement parameter (enlargement / reduction parameter) indicating the expansion or reduction ratio of the chain S, and a chain When the motion parameter (rotation parameter) representing the rotation amount is R, the motion parameter calculation circuit 42 calculates the motion parameters (MVx, MV) according to the following equation, for example.
y), S, and R are calculated.

(MVx, MVy) = ((MVx1 + MVx2) / 2, (MVy1 + MVx2) / 2) S = ((1-F) ² + G ² ) ^1/2 R = arctan (G / (1 -F)) where F = (AC + BD) / E, G = (AD-BC) / E, A = MVx1-MVx2, B = M
Vy1-MVx2, C = x2-x1, D = y2-y1, E = C ² + D ² .

The above equation is obtained by substituting the coordinates of two points and the motion vector of the two points into the equation using the affine parameters shown in FIG.

The motion parameter calculation circuit 12 as described above
According to 5, the following motion parameters are detected.
That is, for example, as shown in FIG. 35 (A), when the linear chain CH _A exists in the current frame,
When the chain is not moving at all in the rear frame, as shown in FIG. 45 (B),
For example, the start point or the end point of the chain CH _A is set as the first or second trust point (in this case, the chain CH _A
Is linear, and therefore the shape reliability calculation circuit 5
No trust candidate points are output from 1, so chain CH _A
The start point or the end point of is the first or second confidence point), and the motion parameter indicating that the chain CH _A is not moving is calculated.

Further, for example, as shown in FIG. 36 (A), when a U-shaped chain CH _B exists in the current frame, the chain is expanded in the rear frame. 36B, for example, each of the two vertices of the U shape of the chain CH _B is set as the first or second confidence point (in this case,
Since the chain CH _B has a U-shape, the shape reliability calculation circuit 51 determines that the points near the two vertices of the U-shape that satisfy the condition described in step S34 are the reliability candidate points. (In the present embodiment, among such reliability candidate points, two vertices in a U-shape are each regarded as the first or second reliability point by the prediction error reliability calculation circuit 52)). As the motion vector of the first or second confidence candidate point, a vector indicating the movement in the lower left direction or the lower right direction is detected. As a result, a motion parameter indicating that the chain CH _B has been expanded is calculated.

As described above, the motion parameter calculation circuit 1
According to 25, the motion parameters that represent the parallel movement, scaling, and rotation that are allowed by the rigid body can be trusted among the points (feature points) that make up the chain. It can be calculated from two points and the motion vector of the two points. Therefore, an accurate motion parameter can be obtained.

Further, the motion vector of the confidence point can be calculated only by obtaining the prediction error while changing the above-mentioned two coefficients Δx and Δy. Therefore, as in the conventional case, the coefficients a to f can be calculated. The calculation amount can be significantly reduced compared to the case where the prediction error is calculated while changing each of the six coefficients, and as a result, the hardware that performs the real-time processing is realized easily and at low cost. It becomes possible to do.

Although the present invention has been described in the case of being applied to an image coding apparatus for performing chain coding, the present invention is not limited to this, and any other image can be coded by detecting the motion of consecutive points in the image. It is applicable to an encoding device. Further, the present invention can be applied to, for example, a video tape recorder, a video conference system, an image editing device, etc., in addition to the image encoding device.

In the present embodiment, the angle reliability and the prediction error reliability are used as the reliability for selecting a reliable point from the points forming the chain. As for, it is possible to use other things (index value) (evaluation amount). That is, for example, it is possible to filter the periphery of the pixel of interest with a high-pass filter and use the output value as the reliability of the pixel of interest. Also, for example, the edge strength of the pixel of interest can be used as its reliability. Furthermore, in this embodiment, two types of reliability, that is, the angle reliability and the prediction error reliability are used together, but the reliability is 1
Only one kind may be used, or three or more kinds may be used.

Further, in the present embodiment, the starting point and the ending point of the chain are not necessarily the candidate points for reliability, but the starting point and the ending point of the chain can always be the candidate points for reliability. Is. Furthermore, it is also possible to always make the start point or the end point of the chain the first or second confidence point, respectively.

Further, in the present embodiment, the motion parameter of the chain to the rear frame is detected, but to the previous frame (the frame one frame before the frame in which the chain for which the motion parameter is to be obtained exists). It is also possible to detect the motion parameter of the. This may be done by storing the image data of the previous frame instead of the subsequent frame data at the place where the subsequent frame data is stored in the above description.

Further, in the present embodiment, two of the points composing the chain are selected and the motion parameter of the chain is calculated. However, the number of points is not limited to two, and for example, one or three. It is also possible to select the above and calculate the motion parameter of the chain. 1
To select only points, the prediction error reliability calculation circuit 52
Then, from among the reliability candidate points, the one with the highest prediction error reliability may be selected. In addition, in order to select three or more points, the prediction candidate reliability reliability calculation circuit 52
The prediction error reliability may be selected in descending order. By doing so, the confidence point calculation circuit 41
From the (prediction error reliability calculation circuit 52), one or more points and their motion vectors are output. In this case, the motion parameter calculation circuit 42 outputs, for example, the average value of the motion vectors. , Can be calculated as a motion parameter indicating the parallel movement amount of the chain.

[0174]

As described above, according to the motion parameter detecting apparatus and the motion parameter detecting method of the present invention, one or more points are selected from the continuous points of the image, and the motion vector for each of the one or more points is selected. Is detected. Then, the motion parameters of the continuous points are calculated based on the motion vector. Therefore, the accurate motion parameters are
It is possible to calculate with a small calculation amount.

According to the image coding apparatus of the present invention, one or more points are selected from the characteristic points forming the chain, and the motion vector for each of the one or more points is detected. Then, the motion parameter of the chain is calculated based on the motion vector. Therefore, it is possible to calculate an accurate motion parameter with a small amount of calculation, and as a result, it is possible to speed up the encoding process.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment of an image encoding device to which the present invention has been applied.

2 is a block diagram showing a configuration example of a chain encoding circuit 12 in FIG.

FIG. 3 is a diagram for explaining storage contents of a coefficient frame buffer 21 of FIG.

FIG. 4 is a flowchart illustrating an operation of the direction searcher 23 of FIG.

FIG. 5 is a flowchart that follows the flowchart of FIG.

FIG. 6 is a diagram showing storage contents of ROMs 30 and 31 of FIG.

FIG. 7 is a diagram for explaining an address output by the address generator 32 of FIG.

FIG. 8 is a diagram showing stored contents of a ROM 33 of FIG.

9 is a diagram showing a direction represented by direction data output from a ROM 33 of FIG.

10 is a flowchart following the flowchart of FIG.

11 is a timing chart showing the output timing of the valid data selection signal from the direction searcher 23, the direction data from the ROM 33, and the forward direction data from the latch circuit 34 in FIG.

12 is a diagram for explaining direction change data output from the direction change signal generator 35 of FIG.

13 is a diagram showing a correspondence table of forward direction data and direction data stored in a direction change signal generator 35 of FIG. 2 and direction change data.

14 is a diagram showing a correspondence table of forward direction data and direction data stored in the direction change signal generator 35 of FIG. 2 and direction change data.

15 is a diagram showing code words assigned to the direction change data output from the direction change signal generator 35 of FIG.

16 is a diagram for explaining the data output from the selector 36 of FIG.

17 is a block diagram showing a configuration example of a chain replacement circuit 15 of FIG.

18 is a diagram showing stored contents of a RAM 123 _n shown in FIG.

FIG. 19 is a flowchart illustrating the operation of the replacement circuit 124 of FIG.

FIG. 20 is a diagram for explaining the process of step S124 of FIG.

21 is a diagram for explaining multiplexed chain encoded data output from the select multiplex circuit 16 of FIG. 1. FIG.

FIG. 22 is a block diagram showing a configuration of a second embodiment of an image encoding device to which the present invention has been applied.

23 is a block diagram showing a configuration example of a chain replacement circuit 215 of FIG.

FIG. 24 is a flowchart illustrating an operation of the replacement circuit 132 of FIG. 23.

FIG. 25 is a flowchart illustrating details of the process of step S132 of FIG. 24.

FIG. 26 is a diagram for explaining multiplexed chain encoded data output from the select multiplex circuit 216 of FIG. 22.

27 is a diagram for explaining a decoded image obtained by decoding the multiplexed chain encoded data output from the image encoding device in FIG. 1 (FIG. 22).

28 is a diagram for describing a decoded image obtained by decoding the multiplexed chain encoded data output from the image encoding device in FIG. 1 (FIG. 22).

29 is a block diagram showing a configuration example of the motion parameter calculation circuit 125 of FIG. 17 (FIG. 23).

30 is a block diagram showing a configuration example of a confidence point calculation circuit 41 in FIG.

FIG. 31 is a flowchart illustrating an operation of the shape reliability calculation circuit 51 of FIG. 30.

32 is a diagram for explaining the process of step S33 in FIG. 31. FIG.

FIG. 33 is a diagram for explaining the process of step S34 of FIG.

FIG. 34 is a flowchart illustrating an operation of the prediction error reliability calculation circuit 52 of FIG. 30.

FIG. 35 is a diagram for explaining motion parameters calculated by the motion parameter calculation circuit 125 of FIG. 17 (FIG. 23).

36 is a diagram for explaining motion parameters calculated by the motion parameter calculation circuit 125 of FIG. 17 (FIG. 23).

[Fig. 37] Fig. 37 is a block diagram illustrating a configuration of an example of a conventional image encoding device.

FIG. 38 is a block diagram showing a configuration of an example of a conventional motion parameter detection device.

FIG. 39 is a coordinate conversion formula (Mo
It is a figure which shows the general expression format of (tion Vector Transform Equation).

FIG. 40 is a diagram for explaining motion parameters obtained by the motion parameter detection device of FIG. 38.

41 is a diagram for explaining motion parameters obtained by the motion parameter detection device of FIG. 38. FIG.

[Explanation of symbols]

10 Two-dimensional change point detection circuit 11 Quantizer 12 Chain coding circuit 15 Chain replacement circuit 16 Select multiplexing circuit 41 Confidence point calculation circuit 42 Motion parameter calculation circuit 51 Shape reliability calculation circuit 52 Prediction error reliability calculation circuit 53 Read Circuit 54 Current frame buffer 55 Rear frame buffer 121 Chain map circuit 124 Replacement circuit 125 Motion vector calculation circuit 131 _{1 to} 131 _N-1 New chain coding circuit 132 Replacement circuit 133 _{1 to} 133 _N-1 Chain decoding circuit 215 Chain replacement Circuit 216 select multiplex circuit

Claims (12)

[Claims] 1. A motion parameter detection device for detecting a motion parameter, which is a parameter representing the motion of continuous points in an image, comprising: selecting means for selecting at least one of the continuous points; and selecting means. Motion vector detecting means for detecting a motion vector for each of the one or more points selected by: and a motion parameter for calculating a motion parameter of the continuous point based on the motion vector detected by the motion vector detecting means. A motion parameter detecting device comprising: a calculating unit. 2. The selecting means is configured to select two of the continuous points.
A point is selected, and the motion parameter calculation means calculates the motion parameter indicating the parallel movement amount, the enlargement / reduction ratio, or the rotation amount of the continuous point based on the motion vector for each of the two points. The motion parameter detection device according to claim 1. 3. The selecting means calculates a reliability representing reliability of each of the continuous points, and based on the reliability,
The motion parameter detection device according to claim 2, wherein the two points are selected. 4. The selecting means calculates the reliability of each of the continuous points based on a prediction error obtained by performing block matching using blocks in a predetermined range centered on the continuous point. The motion parameter detection device according to claim 3, wherein 5. The motion parameter according to claim 3, wherein the selecting unit calculates the reliability of each of the continuous points based on an angle formed by the continuous points with the point as an apex. Detection device. 6. The selecting means selects one of the continuous points that satisfies a predetermined condition and has the highest reliability and the next highest point. 3. The motion parameter detection device according to item 3. 7. The selecting means, when two or more points satisfying the predetermined condition do not exist among the continuous points,
The motion parameter detection device according to claim 3, wherein a start point and an end point of the continuous points are selected. 8. The selection means selects a start point and an end point of the continuous points.
The motion parameter detection device described in 1. 9. The motion parameter calculating means calculates an average value of the motion vectors detected by the motion vector detecting means as the motion parameter representing the parallel movement amount of the continuous points. 1. The motion parameter detection device according to 1. 10. The selecting means calculates a reliability representing the reliability of each of the continuous points, and selects one or more of the continuous points based on the reliability. The motion parameter detection device according to claim 9. 11. A motion parameter detection method for detecting a motion parameter, which is a parameter representing a motion of continuous points in an image, wherein one or more points among the continuous points are selected, and each of the one or more points is selected. Is detected, and the motion parameter of the continuous point is calculated based on the motion vector. 12. A feature point coding unit that encodes information about a feature point of a moving image and outputs coded data, and a motion parameter detection that detects a motion parameter that is a parameter representing a motion of a chain connecting the feature points. Means, encoded data output from the feature point encoding means,
An image coding apparatus comprising: a multiplexing unit that multiplexes the motion parameter detected by the motion parameter detecting unit, wherein the motion parameter detecting unit is one of the feature points forming the chain. Based on the selection means for selecting the above points, the motion vector detection means for detecting the motion vector for each of the one or more points selected by the selection means, and the motion vector detected by the motion vector detection means And a motion parameter calculation unit that calculates a motion parameter of the chain.